中文English

Lottery, Salary or Startup: The Odds and Costs of Three Roads to NT$100 Million

JIAHON Group · JIAHON Think Tank perspective · Road to NT$100 Million, Part One. Published by JIAHON Think Tank | Written by Marcus (Senior Vice President) | Published 20 September 2026 | Updated 2 October 2026 | jiahon.com.tw

Taiwan has about 8,000 people with net assets above NT$1 billion; lottery winners of NT$1 billion or more since the games began number fewer than twenty. Eight thousand to twenty — founding a company looks far better. This article takes that intuition apart all the way down. The same number was revised nine times, each time because the previous version hid an assumption nobody had checked. The conclusion runs against the intuition, and the process of taking it apart is worth more than the conclusion.

In Brief

Taiwan has roughly 8,000 people with net assets above NT$1 billion and fewer than twenty lottery winners at that level since 2007. The ratio makes founding a company look like the obviously better bet. This note tests that intuition on one ruler—the lottery, whose price per unit of “NT$100 million probability” is public and linear (about NT$1.04 billion of wagers per NT$100-million winner)—and then measures the other two roads against it. The entrepreneurship estimate had to be revised nine times, from 308× the lottery’s odds down to a ladder of 21–77×, each revision removing a hidden assumption: the wrong denominator, survival (24% of new firms die in year one, 44% remain at ten years), the cumulative-revenue trap, and the fact that net margin is a function of gross margin rather than a number to guess.

Two findings carry the piece. First, for the average entrant none of the three roads has a positive expected value; the real choice is which shape of loss you can afford. Second, the split inside entrepreneurship—commodity wholesale versus a niche technical channel—moves the odds 3.7×, which is larger than the gap between entrepreneurship and the alternatives. One thing flips the sign inside the model: a ten-year industry “ticket” acquired before founding. A second—growth of the surviving company—is the model’s open cell, and the subject of the sequel. The pattern is not unique to Taiwan: in U.S. data, the median entrepreneur ten years into a business earns 35% less than a comparable employee (Hamilton, 2000), and roughly one in five new establishments with employees closes within its first year (U.S. Bureau of Labor Statistics).

1. Eight thousand to twenty

Start with two numbers. The Taiwan Ultra-High-Net-Worth Wealth Insights Report released by CTBC and BCG in January 2026 estimates that about 8,000 people in Taiwan have net assets of NT$1 billion or more, rising to 11,000 by 2029. On the other side, since Taiwan Lottery began in 2007, single winning tickets worth NT$1 billion or more total fewer than twenty.

Eight thousand to twenty. Founding a company looks clearly the better deal.

This article takes that intuition apart all the way down. Three roads go on the same ruler: buying lottery tickets, working and saving, and starting a company. The expected value of the three roads: for the average entrant, none of them is positive. The difference is the shape of the negative — one is small, repeatable and capped; one cannot reach the goal even with perfect execution; one is large, one-shot and uncapped.

The choice is not “pick the road that wins.” It is “pick the kind of loss you can afford.”

And the entrepreneurship side forks once more. Two people both start companies; one does commodity wholesale, the other a niche technical channel, and their odds differ by 3.7×. That fork matters more than the decision whether to start a company at all.

2. The lottery as a ruler: the public price of a winning chance

To compare roads to wealth you need a ruler first. The lottery is a good one, because it has four properties nothing else has: the price is public, it is fully linear, supply is unlimited, and anyone can buy.

Taiwan Lottery reported that in 2025 the NT$100-million-plus jackpots produced 22 winning tickets: 17 for Lotto 6/49 and 5 for Super Lotto. Back out the betting volume from the jackpot odds: Lotto 6/49 has odds of 1 in 13,983,816, so 17 jackpots need about NT$11.9 billion of wagers; Super Lotto has odds of 1 in 22,085,448, so 5 jackpots need about NT$11.0 billion. Together that is about NT$22.9 billion — every NT$1.042 billion or so of wagers produces one NT$100-million winner. That is the public price of the product called “a chance at NT$100 million.”

Table 1: Converting at the lottery price — annual spending and the chance of NT$100 million

Annual spendingChance of NT$100 millionChance of NT$1 billion
NT$1,200 (national average)1 in 870,0001 in 16 million
NT$10,0001 in 104,0001 in 1.93 million
NT$100,0001 in 10,4001 in 190,000

Incidentally, the average adult spends only about NT$1,187 a year on Lotto 6/49 and Super Lotto. The idea that “everyone is buying lottery tickets” does not hold — that is about NT$100 a month of entertainment, not a road to wealth.

This ruler gives one test: for any road to wealth, whatever it costs per unit of NT$100-million probability above NT$1.042 billion is money paid for something other than money.

3. The entrepreneurship side took nine measurements to get right

Entrepreneurship has no public price; you have to calculate it. This chain of calculation was revised nine times, and each revision removed a hidden assumption.

Figure 1: The same number, revised nine times — from 308× to a ladder of 21–77×

Three key revisions, one sentence each. The denominator was wrong: “8,000 new NT$100-million fortunes a year” divided by “26 NT$100-million lottery winners a year” gives 308×, but those two groups were not drawn from the same pool; the right pool uses business owners as the denominator, narrowed to the 768,408 firms in wholesale and retail. Time was left out: once you include whether a new company is still alive ten years later, the multiple drops from 296 to 110; and accumulating net worth needs “ten years of cumulative revenue,” not “revenue in year ten” — high growth is itself the enemy of accumulation, because an exponentially growing company contributes very little in its early years. Net margin should not be guessed: the first eight versions treated net margin as a number to assume; the Ministry of Finance’s industry profit standards give both gross and net margin, and a regression across 1,283 categories shows net margin is a function of gross margin. That cell moved from assumption to lookup, and it split the final version into a ladder — Section 7 comes back to it.

This error has a name: survivorship bias. During the Second World War, the statistician Abraham Wald analysed bullet holes in returning bombers and pointed out that the places with few holes were the vital ones — planes hit there never came back, so they were never counted. The 8,000 billionaires are the planes that came back; the ones that did not will never appear in any wealth report. Taleb, in The Black Swan, calls this “silent evidence.” Section 4 brings the silent group back.

4. The real survival rate of Taiwanese startups: 24% gone in year one, 44% left at ten years

The cut in year one is more than four times heavier than in any of the following nine years

Public statistical tables cannot give you a survival rate. The Ministry of Economic Affairs’ company registration statistics have only three columns: existing, newly established, and dissolved/revoked/annulled. In 2025 there were 806,188 existing companies, 42,752 new ones and 28,064 dissolved, revoked or annulled — a legal death rate of 3.48% a year, implying a ten-year survival rate of 70.2%. But that number is false, because most failed companies never file for dissolution; they simply stop issuing invoices.

The real data sit in two firm-by-firm open datasets from the Ministry of Finance: the national business (tax) registration lists 115,067 businesses as “suspended” and 2,062,370 as “not operating, other than suspended.” The second figure, 2.06 million, is larger than the number of small and medium enterprises operating in Taiwan today (1.716 million) — it is a seventy-year cemetery and cannot serve as a denominator. But both files carry an establishment date, so they can be aligned by founding year with the white-paper annex tables to calculate true cohort survival. The census of industry and services, which the Directorate-General of Budget, Accounting and Statistics (DGBAS) carries out on site every five years, then calibrates the 3%–8% of businesses that “have a tax registration but are not actually operating,” giving the operating-basis curve:

Figure 2: The real survival curve — solid line on an operating basis (used), dashed line on a tax-registration basis

24.0% are gone in the first year, and 44.0% survive ten years. The cut in year one is more than four times heavier than in any of the following nine years.

These two numbers also correct the widely shared claim that “90% fail in year one and 1% remain after five years”: 24% fail in the first year, not 90%, and 44% are still there after ten. They also hold up against international data. The U.S. Bureau of Labor Statistics followed a whole cohort of establishments that opened in March 1994: 79.6% were still open after one year, 49.6% after five and 33.6% after ten. The bases are not identical — the U.S. counts establishments with employees, not companies, and excludes one-person businesses — but the order of magnitude matches: Taiwan loses slightly more in year one than the U.S., and keeps more after ten years.

The bar for “survival” is set too low, so survival is not a proxy for success

A company that still issues invoices counts as surviving. Of the 768,408 wholesale and retail businesses in Taiwan, more than half (51.2%) have annual revenue below NT$2 million — under NT$170,000 a month; 22.2% are at NT$500,000 or less, not enough to pay one person’s salary. Statistically, these companies survive. As we will see, the median annual income of owners of ordinary surviving companies is about NT$1.75 million, while the break-even bar for starting a company is NT$3.55 million. Any statistic that uses “still operating” as the measure of success will systematically overstate the success rate.

Ask next “how many businesses are actually operating in Taiwan,” and the answer depends on the basis — and the bases differ by a factor of two: 3.91 million ever registered, 1.735 million operating on tax records, 1.467 million confirmed on site by the census, 841,000 with annual revenue of NT$2 million or more, and 467,000 that actually employ people. That last 467,000 plus 1.270 million own-account workers equals 1.737 million — only 0.1% off the Ministry of Finance’s 1.735 million on tax records. Two agencies, two methods, confirming each other, and at the same time taking that 1.735 million apart: of Taiwan’s 1.735 million “businesses,” only 467,000 actually employ people; the other 1.27 million are one-person businesses.

5. The control group: can a salaried worker save NT$100 million?

The road most people actually take is: work, save, invest. To accumulate NT$100 million of net worth over a forty-year career, what salary do you need? At a 5% real return and a 30% savings rate, you need NT$2.76 million a year — and that assumes the salary is that number from the first year and stays flat for forty years. A real salary is a curve, low early, higher later, with a ceiling: NT$550,000 at 25, NT$1.12 million at 35, NT$1.63 million at 45, then flat around NT$1.8 million. At 25 the gap is fivefold, and those are exactly the years when compounding does the most.

Table 2: The salaried road — recalculated on a real salary curve, adding reality layer by layer

ScenarioInvestment accountTotalShare of goal
① Perfect: 40 years uninterrupted, +5% every year, nothing withdrawnNT$44.11 millionNT$44.11 million44%
② Add life’s withdrawals (marriage, home, cars, children’s education, care for parents), with home equity added backNT$0.75 millionNT$27.91 million28%
③ Add four market crashesNT$0.25 millionNT$27.41 million27%

Even the perfect scenario reaches only 44%. In row ② the investment account drops from NT$44.11 million to NT$0.75 million, but it does not go to zero — the house is an asset. A home bought at 35 has net equity of about NT$27.16 million at 65 once the mortgage is paid off; adding that back gives NT$27.91 million. So buying a home is not why this road falls short; it falls short because its ceiling was around NT$40 million to begin with.

Two things cannot be assumed away. Crashes are not a risk; they are a certainty: since 1990 there have been 8 major corrections in 36 years, on average one every 4.6 years; 5 of them fell more than 30%, on average every 8.0 years. A forty-year career will meet five or six corrections of more than 30%, and the losses are asymmetric — a −35% at age 55 takes NT$6.15 million in one go, because the balance is larger by then. That money cannot be touched for forty years: row ① holds only if marriage, buying a home, changing cars, having children and a parent falling ill all never happen.

There is a steeper detail: where do you have to stand to earn NT$2.76 million a year? Across all 1,823 listed and OTC companies, the median of their “median employee salary” is NT$734,000. At one large EMS group, non-managerial full-time employees average NT$2.768 million with a median of NT$2.028 million (ranked 27th of 1,823, and excluding production-line staff). DGBAS puts the median total annual pay of all employees at NT$546,000, NT$1.099 million for those with a graduate degree, and NT$1.106 million in finance and insurance, the highest-paying industry. “A salary of NT$2.76 million” is exactly the average of a few thousand people at that EMS group, who have already been filtered once.

Working plus investing does not reach NT$100 million: 44% with perfect execution, 28% with a normal life. It is not the easier of the three roads; it is the only one that “cannot reach the goal even with perfect execution.” But the lesson to take from here is not “a job is useless” — it supports a life, builds human capital and carries lower risk; it just does not lead to NT$100 million. If the goal was never NT$100 million, this section does not apply, and that may well be the wiser goal.

6. Three ways to lose: first ask whether you can try again, then look at expected value

Up to here we have compared probabilities. But a decision has to pass two gates, in an order that cannot be reversed: first ask whether you can try again after losing, then look at expected value. Expected value is an average spread across a group of people. In a game where you bet a fixed amount each time and can try again after losing, one person playing long enough moves toward that average; when the stake rolls along with wins and losses, they do not — one wipe-out and there is no next round, and however good the average is, you never get to collect it. The physicist Ole Peters argued in Nature Physics in 2019 that economics has long treated these two averages as the same thing; Taleb says it more plainly in Skin in the Game: a hundred people each going to a casino once, and one person going a hundred times, are two different things. First lay out the expected value — both roads are negative; whether you can try again is answered by the recovery rates at the end of this section. Over twenty years, the baseline is “keep working” (salary from NT$2 million, twenty years of pay totalling NT$42 million, a model setting):

Table 3: The full distribution of twenty years of entrepreneurship (baseline: keep working, NT$42 million)

BranchPersonal wealth after twenty years19% step (21×)57% step (67×)
Reach NT$100 millionNT$120 million0.202% → NT$0.24 million0.643% → NT$0.77 million
Survive but stay below NT$100 millionNT$35 million43.8% → NT$15.33 million43.4% → NT$15.17 million
FailNT$18.58 million56.0% → NT$10.41 million56.0% → NT$10.41 million
Expected value of entrepreneurship—NT$25.98 millionNT$26.35 million
Net difference (vs keep working, NT$42 million)—−NT$16.02 million−NT$15.65 million

The two columns are the 19% and 57% gross-margin steps; the −NT$15.57 million in Section 7 is the 65% step.

The NT$18.58 million on failure breaks down as follows: the company lasts about 3.54 years on average before closing, at NT$1 million a year, NT$3.54 million in total; NT$8 million is lost (NT$10 million invested, a 20% recovery rate, only NT$2 million back); the remaining 16.46 years are spent re-employed at a 30% discount, NT$1.4 million a year, NT$23.04 million in total.

A negative expected value is not peculiar to Taiwan. Barton Hamilton compared entrepreneurs and employees on U.S. data in the Journal of Political Economy in 2000: entrepreneurs ten years into their business have median earnings 35% below comparable employees, and the gap cannot be explained by “less able people are the ones who start businesses.” Tobias Moskowitz and Annette Vissing-Jørgensen found in the American Economic Review in 2002 that entrepreneurs concentrate their wealth in their own company, bearing far more risk than buying the stock market, without a higher return. Both point to the same puzzle: the money does not add up, and people do it anyway. Taiwan’s answer is in Section 9.

One point stands out here: the kind of business you choose can widen the chance of reaching NT$100 million by more than three times, yet the expected value barely moves. The reason is that the expected value is dominated by the 43% that “survive but stay below NT$100 million”; the NT$100-million cell carries too little weight. The kind of business decides the tail, not the expected value.

Table 4: The three roads side by side — the real choice

LotteryWork + investStart a company
Stake per attemptNT$10,000 a year30% of salaryNT$10 million of capital + a twenty-year career
Repeatable?Yes, without limitYes, once a yearOnce; hard to try again after failing
Loss capYes, the amount stakedYes, the principalNone; personal guarantees reach through
Result of perfect executionStill negativeNT$44.11 million (short of goal)Decided by execution
Expected valueFixed at −45%Positive, but does not reach the goalNegative, but can be changed by execution
Chance of NT$100 million0.0096%Very low0.202%–0.643%

The lottery “looks good” purely because it is small: on NT$100,000 staked, the loss ceiling is right there. Starting a company stakes NT$10 million of capital plus a twenty-year career, a loss space two orders of magnitude larger. So the real question is not “which road wins” but “which way of losing do you want” — the lottery loses NT$4,500 a year and can sit there as a background constant; a job has a positive expected value and lower risk, but its ceiling is around NT$40 million; a company has a negative expected value and no floor on the downside, but it is the only one of the three whose expected value is not fixed — that minus sign can be changed by execution. The expected values of the first two roads are parameters of the game; the expected value of the third is written by you. That is the only real difference in expected value.

How much you get back on failure

Figure 3: How much you get back on failure — opening a restaurant (8%) is worse than buying lottery tickets (55%)

The recovery rate is what really decides the size of a loss on failure. In a restaurant, fit-out labour is 40% to 60% of the investment, and the recovery rate on that part is not merely low, it is structurally zero — it was never an asset, it was a one-time service purchase; the refrigerator can be sold second-hand, the carpenter’s hours cannot. The 20% in a wholesale channel is not because the stock cannot be sold; it is the order of liquidation: wages and severance, taxes, secured bank debt, supplier payables — after those four in line, what is left for shareholders is usually close to zero.

One thing needs to be clear: the recovery rate decides the risk of ruin, not the expected value. Its coefficient in the expected-value formula is only 0.043, but it decides whether you can try again after failing. Fail in a business with a 60% recovery rate and you still have 40% of your capital; fail in one with an 8% recovery rate and you are at zero and in debt. The same expected value; one is repeatable, the other is not.

That is the gate at the start of this section: expected value answers how a group does on average; the recovery rate answers whether the same person survives to play the next round. In October 1998 at the University of Florida, speaking about Long-Term Capital Management (LTCM), which had just come close to collapse and been taken over with capital from a bank consortium, Warren Buffett said the same thing: to make money you do not have and do not need, risking what you do have and do need is not worth it, however good the odds.

7. The real fork is not whether to start a company but what kind of business: the gross-margin ladder

So far we have compared “entrepreneurship versus everything else.” But once the official profit standards are pulled in, the gap inside entrepreneurship turns out to be larger than the gap between entrepreneurship and the other options. The Ministry of Finance’s 2025 (ROC year 114) Industry Income and Profit Standards for Profit-Seeking Enterprises contains 1,283 parsable tax industry categories; within wholesale, the 47 categories for industrial materials and equipment have a median net margin of 8.0% and a median gross margin of 19.0%. A niche technical channel — anti-static materials, cleanroom consumables, SMT-related items, things that must first be measured, confirmed and planned — can reach three times the official typical gross margin (industry estimate). Both sit under “wholesale,” and they are two completely different businesses.

Net margin is not a parameter to guess

That book gives both gross and net margin, and a regression across the 1,283 categories gives: net margin = 0.348 × gross margin + 1.51 (r = 0.897, residual standard deviation 2.83 percentage points). The group medians agree: categories with 10–20% gross margin have a median net margin of 8%; 20–30%, 9%; 30–40%, 12%; 40–50%, 14%; 50–60%, 18%; 60–75%, 26%. So net margin is a function of gross margin, and gross margin can be measured — this cell moves from assumption to lookup.

Figure 4: The gross-margin ladder — two people start companies, and their odds differ by 3.7×

Table 5: The gross-margin ladder — the bar at each step and how many firms clear it

Gross marginNet marginYear-10 revenue barFirms clearing itMultiple
19% (official typical value for wholesale)8.1%NT$707 million1,83121×
30%11.9%NT$480 million2,90733×
40%15.4%NT$372 million3,94845×
50%18.9%NT$304 million5,03658×
57% (reachable for a niche technical channel)21.3%NT$269 million5,82267×
65%24.1%NT$238 million6,74377×

Put the ladder back into the twenty-year expected value: the net difference is −NT$16.02 million at the 19% step and −NT$15.57 million at the 65% step — a 3.7× difference in the multiple, but only NT$450,000 in expected value (about 2.8% of the net difference). Climbing the margin ladder does not make starting a company a good deal; it makes the outcome “in case it works” more likely. Gross margin decides how thick the tail is, not the expected value.

That NT$450,000 is a lower bound. The model fixes wealth after reaching NT$100 million at NT$120 million, but firm size is fat-tailed by nature: Robert Axtell, in Science in 2001, used data on all U.S. tax-filing firms to show that firm sizes follow a Zipf distribution with an exponent close to 1; the tail index of 1.197 that this article backs out from the entry thresholds of the China Credit Information Service rankings falls in the same range. With that index, a wealth cap of NT$500 million gives an expected-value gap of NT$850,000; with no cap, NT$3.08 million. The sign still does not flip, and margin still mainly changes the tail — only its effect on expected value is more than NT$450,000.

Gross margin is not given by the market. Anyone can enter the 19% step, so nobody makes money there; the 65% step requires customers to write you into their specifications, and that takes ten years. This margin ladder is really the projection of another ladder — how it lines up with the pillar order in “Four Pillars” is the subject of Part Two.

One limitation must be stated: the industry profit standards are not a measurement of actual profitability; they are a lower bound set for tax assessment, and trade associations have a motive to push them down. So the whole ladder should be read as a conservative lower bound, and the regression relationship (the slope from gross to net margin) is more reliable than the absolute level. Sampling five uncertain parameters within their reasonable ranges and running 200,000 simulations: commodity wholesale (19% gross margin) has a median multiple of 28×, 80% interval 13–48; a niche technical channel (57% gross margin) has 87×, 80% interval 59–124. The variance comes from three cells: the accumulation coefficient 39%, the net-margin residual 30% and the take-home rate 27%, 96% together; the tail index and survival rate, which this article spent the most effort calibrating, together account for only 4%. The parameter that is easy to calibrate is usually the one that does not matter.

8. What can turn the expected value of starting a company positive: the industry ticket

The ticket cannot be divided, cannot be transferred and expires — you can only grow it yourself over ten years

The expected value comes out negative, but the break-even bar is not fixed; it rises linearly with opportunity cost. For an ordinary employee at NT$720,000 (the opportunity cost set in the model; the NT$546,000 in Section 5 is the DGBAS median total annual pay, on a different basis), the break-even bar is NT$900,000; for the graduate-degree average of NT$1.13 million (an average; the NT$1.099 million in Section 5 is a median), it is NT$1.75 million; for a senior tech employee at NT$2 million, NT$3.55 million. The owner of an ordinary surviving company earns about NT$1.75 million a year — that horizontal line lands right on the graduate-degree point: a person with a graduate degree who starts a company roughly breaks even, no more, no less.

So what can push the expected value positive? The answer is barriers to entry. And here is the counter-intuitive part: a high barrier does not make starting a company easier; it moves the hard part to before you start. Compare two paths, both twenty years, removing the NT$100-million branch, which is only 0.9%–3.4% of the expected value (across the 19% to 65% margin steps):

Figure 5: The price of the ticket — A loses to employment however you count it, B beats it; the cost is only half the years of running the business

A: starts a company at thirty and runs it for twenty years. Survival 44.0% × NT$35 million, failure 56.0% × NT$18.58 million; expected value NT$25.81 million, a net difference of −NT$16.19 million against the NT$42 million baseline. B: works in the industry from thirty to forty, building the ticket on a salary, starts a company at forty and runs it for ten years. The first ten years’ salary of NT$20 million is certain; in the last ten years, survival 65% × NT$40 million, failure 35% × NT$4.58 million; expected value NT$47.60 million, a net difference of +NT$5.60 million. The ticket is worth NT$21.80 million. The trade pays because survival rate and margin multiply, while years of operation add.

This flip needs two numbers to hold at the same time: 65% survival with a ticket, and NT$4 million of annual income. Raise only survival to 65% with income still at NT$1.75 million, and the net difference is −NT$9.02 million; raise only income to NT$4 million with survival still at 44%, and it is −NT$1.84 million; only with both is it +NT$5.60 million. Of the two, income carries more weight.

Data from abroad stand on B’s side. Pierre Azoulay and colleagues, in American Economic Review: Insights in 2020, studied new firms using U.S. Census Bureau administrative data: founders of the top 0.1% fastest-growing new firms were on average 45 when they started, and prior experience in the same industry greatly raised the chance of success. Rajshree Agarwal and colleagues, studying the disk-drive industry in 2004, also found that companies founded by employees who left a parent firm lived longer than other entrants; the stronger the parent’s capabilities, the stronger the capabilities the new firm inherited, and higher technical capability meant higher survival — the knowledge came along with the people.

Three properties of the ticket

It expires. Industry relationships decay — buyers are reassigned, engineers retire, processes move to a new generation. The network that works at forty is less than half as useful at fifty-five; the ticket has not only a price but an ideal time to exercise it. It cannot be divided or transferred. You cannot buy 30% of an industry track record, and you cannot let someone else use your ticket; you can only grow it yourself over ten years. And this is the deeper reason the recovery rate is only 20%. Most of the value of a B2B distributor is its principal’s ticket, and the ticket does not transfer with the assets — a buyer gets the stock, the receivables and the customer list, but not “customers are willing to take your call.” The philosopher Michael Polanyi described this tacit knowledge in one line in The Tacit Dimension in 1966: we know more than we can tell.

That gives an uncomfortable structure: company valuation = transferable assets (high recovery rate) + the principal’s ticket (zero recovery rate). The larger the ticket’s share, the more profitable the business, and the harder it is to sell. Until the ticket is written down, this structure has no solution — Part Two explains how it is solved.

9. Those with tickets don’t enter; those who enter have no ticket

All the numbers have been calculated, but the real question is: if the expected value is negative, why do so many people do it? Taiwan sets up 108,531 new small and medium enterprises a year, 0.56% of the adult population; at that rate, the chance that a person starts a business at some point in a forty-year working life is about 20% — one in five, among the highest densities of entrepreneurship in the world. The lottery, meanwhile, gets only NT$1,187 a year per person. The road with the far more negative expected value has the higher participation. Standard utility theory cannot explain that, but once the ticket and opportunity cost are included, it makes complete sense.

Behavioural economics has offered another explanation. Colin Camerer and Dan Lovallo reported experiments in the American Economic Review in 1999 showing that when success depends on skill, more people enter than should: everyone knows most will fail, but forgets that those willing to enter are the ones who think they are better than average — what they called “reference group neglect.” The explanation in this article does not conflict with it; it is more specific: in Taiwan, the people who should enter are the ones who don’t.

Table 6: For a ticket to count, three things must hold together — scarce, portable, redeemable

PositionWhat you learnScarcePortableRedeemable
Production-line operatorStandard work✗✓✗
Restaurant floor staffService flow✗✓✗
Delivery driverRoutes✗✓✗
Process engineerWhy things fail, how to match specs✓✓✓
Sales managerWho decides, how to negotiate✓Partly✓

The common structure of low-paid positions is that they touch “execution,” not “decisions.” And the essence of a ticket is “knowing how other people decide” — why the buyer picked this supplier, why the engineer is holding this specification, why the boss cut this budget. An operator is on the shop floor every day, but the shop floor does not produce that information. So it is not a question of ability; it is a question of position. Put the same person in a process engineer’s seat and in ten years they have a ticket; put them on the line and in ten years they have only seniority. A harder inference follows: low pay is not only low pay; it also means “no ticket in the future either” — low pay is a current loss, while not building a ticket is a compounding loss, and the latter costs more.

Why people with higher education do not enter needs opportunity cost split into three layers. Cash: the NT$3.55 million break-even bar. Certainty: a NT$2 million salary is not just NT$2 million; it is a certain stream arriving every year for twenty years without bearing risk, and swapping it for a bet with a 56.0% chance of going to zero requires compensation far above the paper difference. Sunk identity cost: someone with less education who fails goes back to the old job; someone with more education who fails cannot go back to the old seat. Add the three layers, and not entering is the right answer.

And there are direct measured data on this. Among people employed in small and medium enterprises, 36.99% have a university degree or higher; but within the same group, only 28.61% of employers have a degree, and only 18.72% of own-account workers (46.68% of employers at large enterprises have degrees, but they are professional managers after scaling, not a group that started from nothing). The higher the education, the less likely a person is to become their own boss — that is not an inference; it is measured. There is also a signal in age: in 2024, 39.59% of SME employers were 55 or older (37.85% the year before), and only 27.72% were under 45. The employer population is ageing, and young people are not coming in.

Taiwan’s high participation and low average quality in entrepreneurship are not a matter of preference; they are a matter of allocation. Those with tickets don’t enter; those who enter have no ticket; so the industries that surface are the ones that hardly need a ticket. This explains three scattered facts: 57.8% of SMEs are sole proprietorships, more than half have annual revenue below NT$2 million, and 24.0% die in the first year — low-barrier entrants are the majority, businesses without a ticket do not grow, and entering without a ticket has an immediate cost.

10. What this means for mid-stream supply chain companies

Survival is not a proxy for success — look at the revenue bracket, not whether it is still open.
When assessing a supplier or a customer, “still in business” tells you nothing — half of wholesale and retail companies have annual revenue below NT$2 million. Look at which bracket it is in and which direction it is moving.

The margin ladder is a choice, not luck — 19% to 65% is a 3.7× difference.
Both are listed as wholesale; at the 19% step customers ask about price, at the 65% step customers write you into their specifications. Each step up corresponds to a change in the question the customer asks; a company can travel this road further than a person, because a company can write the ticket down — that is the subject of Part Two.

The larger the ticket’s share, the more profitable and the harder to sell — turn the ticket into a transferable asset.
Most of the value of a B2B distributor is its principal’s ticket, and the ticket does not transfer with the assets. This is a structure shared by all mid-stream companies, and it is why “making money” and “being worth money” are two different things.

The cell in this model that is still broken

The expected-value model assumes the owner of a surviving company takes NT$1.75 million every year for twenty years, but companies grow. Starting at NT$1 million with 8% annual growth, the net difference shrinks from −NT$15.69 million (the baseline of the eighth version, the 62× model) to −NT$11.02 million; starting at NT$1.2 million with 8% growth, −NT$7.05 million; starting at NT$1 million with 12% growth, +NT$0.39 million — it flips positive. Using the baselines of each step of the final ladder (−NT$15.57 million to −NT$16.02 million), 12% annual growth flips it positive as well. In other words, the conclusion “the expected value of starting a company is negative” rests on an assumption never stated: that surviving companies do not grow. That is where the whole model is fragile, and it is where Part Two begins.

For the average entrant, none of the three roads is positive; the difference is the shape of the loss. The fork inside entrepreneurship — what kind of business — decides the tail. Of the things that can flip the sign, this article found one: the ticket from the ten years before entering. The other cell — the growth rate of the surviving company — is left for Part Two.

The expected values of the first two roads are parameters of the game; the expected value of the third is written by you. How to write it: see Part Two, “You Write the Sign.”

Frequently asked questions

What is the survival rate of startups in Taiwan?

Tracking Ministry of Finance tax-registration data by founding year and calibrating actual operation against the DGBAS census of industry and services, about 24% of new companies stop operating in the first year, and about 44% are still there after ten years. The widely shared claim that “90% fail in year one and 1% remain after five years” is far from the actual data.

What are the odds of winning the lottery jackpot?

The Lotto 6/49 jackpot is 1 in 13,983,816, and the Super Lotto jackpot is 1 in 22,085,448. Backing out from the 22 jackpots of NT$100 million or more drawn in 2025, every NT$1.042 billion or so of wagers produces one NT$100-million winner.

What salary does a worker need to save NT$100 million in forty years?

At a 5% real return and saving 30% a year, you would need a salary of NT$2.76 million from the very first year, unchanged for forty years. Recalculated on a real salary curve, even perfect execution reaches only 44% of NT$100 million.

Is the expected value of starting a company positive or negative?

For the average entrant it is negative: over twenty years, about NT$15.65 million to NT$16.02 million less than twenty more years of employment (NT$42 million in total). The difference is in how you lose: a company’s losses have no cap, because personal guarantees reach through to the individual. For the conditions that can flip the sign, see “You Write the Sign”.

What is survivorship bias?

Seeing only the samples that survived and not the ones that disappeared, and so overstating the success rate. Looking only at the 8,000 billionaires and forgetting the companies that closed is a textbook case of survivorship bias.

Appendix: the data grade of this article

Any article that argues with numbers should state which numbers were measured and which were estimated. This article has 14 hard data items, 4 calibrated, 1 inferred, 6 industry estimates and 6 assumptions. The assumptions all sit on the money side (the salary curve and life withdrawals on the salaried road, and the specific amounts for “what happens on failure and on success” in entrepreneurship). On the probability side, apart from the 60% take-home rate and the 65% survival rate for founders with a ticket, which are industry estimates, most items are measured or calibrated; the money side is almost entirely soft.

Table 7: Data grades (excerpt)

ParameterGradeSource
Jackpot odds, one NT$100-million winner per NT$1.042 billion, 55% prize payout rateHardCombinatorics; Taiwan Lottery 2025 draw statistics and announcements
Number of wholesale and retail firms and revenue brackets, form of organisationHardSME White Paper annex tables 11, 13
Ten-year survival 46.6% (tax-registration basis)HardMinistry of Finance suspended and non-operating files, aligned by founding year with White Paper annex table 16
Ten-year survival 44.0% (operating basis, used)Hard + external calibrationDGBAS census / tax-registration ratio, two census years confirming each other
Average years to failure 3.54Hard + external calibrationCalculated from the survival curve in Section 4
Numbers of employers, own-account workers and employees, and education structureHardSME White Paper tables 5-1-4, 5-1-5
Median salaries (NT$546,000, NT$734,000)HardDGBAS 2024 statistics; Market Observation Post System 2025 disclosures
Industry profit standards, 1,283 categories, gross-to-net margin regressionHardMinistry of Finance 2025 Industry Income and Profit Standards
Frequency of major corrections since 1990HardTaiwan and international market history
Tail index α = 1.197CalibratedSolved from entry thresholds of the China Credit Information Service Top 500 / Top 100
Accumulation coefficientCalibratedCalibrated to industry high-growth trajectories; high-growth firms contribute little early, so the coefficient is far below the White Paper’s average-trajectory 6.95
Age-group correction factor 0.807InferredAverage revenue ratio, White Paper annex table 16
Take-home rate 60%, recovery rate 20%Industry estimateNo public data to verify
Survival 65% and annual income NT$4 million for founders with a ticketIndustry estimateSet from industry experience
Niche technical channel gross margin 57%, 65%Industry estimateSet from industry experience
Salary-age curve, life withdrawals, home equity, 0.7 re-employment discount, NT$120 million wealth after reaching NT$100 million (conservative lower bound, see Section 7), income during the failure periodAssumptionBuilt from medians and age distributions; no data for the rest

Road to NT$100 Million series: Part One, “The Odds and Costs of Three Roads” | Part Two, “You Write the Sign” | The Remedy, “The Window Opens Only for Those with a Floor” | Part Four, “Margin Is the Price of Trust”

References

[1] CTBC, BCG (January 2026), Taiwan Ultra-High-Net-Worth Wealth Insights Report.
[2] Taiwan Lottery, 2025 (ROC 114) jackpot statistics; game rules for Lotto 6/49 and Super Lotto (jackpot odds).
[3] Small and Medium Enterprise and Startup Administration, Ministry of Economic Affairs, White Paper on Small and Medium Enterprises, annex tables 11, 13, 16; tables 5-1-4, 5-1-5; figure 5-1-5.
[4] Administration of Commerce, Ministry of Economic Affairs, company registration statistics (2025).
[5] Ministry of Finance, national business (tax) registration datasets: suspended; not operating, other than suspended.
[6] DGBAS, Census of Industry and Services (2016, 2021); 2024 statistics on median total annual pay of employees.
[7] Ministry of Finance, 2025 (ROC 114) Industry Income and Profit Standards for Profit-Seeking Enterprises.
[8] China Credit Information Service, Top 5000 Largest Corporations in Taiwan, 2024 entry thresholds for the Top 100 and Top 500.
[9] Market Observation Post System, 2025 (ROC 114) disclosures of non-managerial full-time employee salaries at listed and OTC companies.
[10] JIAHON Think Tank (2026), “Four Pillars, Build Your Own Two First”; “You Write the Sign” (Part Two of this article).
[11] Marc Mangel, Francisco J. Samaniego (1984), “Abraham Wald’s Work on Aircraft Survivability,” Journal of the American Statistical Association 79(386): 259–267. https://doi.org/10.1080/01621459.1984.10478038
[12] Nassim Nicholas Taleb (2007), The Black Swan: The Impact of the Highly Improbable, Random House, chapter 8, “Giacomo Casanova’s Unfailing Luck: The Problem of Silent Evidence.”
[13] U.S. Bureau of Labor Statistics (2018), “18 percent of businesses opening in March 1994 remained open as of March 2017,” The Economics Daily. https://www.bls.gov/opub/ted/2018/18-percent-of-businesses-opening-in-march-1994-remained-open-as-of-march-2017.htm
[14] Ole Peters (2019), “The ergodicity problem in economics,” Nature Physics 15(12): 1216–1221. https://doi.org/10.1038/s41567-019-0732-0
[15] Nassim Nicholas Taleb (2018), Skin in the Game: Hidden Asymmetries in Daily Life, Random House, chapter 19, “The Logic of Risk Taking.”
[16] Barton H. Hamilton (2000), “Does Entrepreneurship Pay? An Empirical Analysis of the Returns to Self-Employment,” Journal of Political Economy 108(3): 604–631. https://doi.org/10.1086/262131
[17] Tobias J. Moskowitz, Annette Vissing-Jørgensen (2002), “The Returns to Entrepreneurial Investment: A Private Equity Premium Puzzle?,” American Economic Review 92(4): 745–778. https://doi.org/10.1257/00028280260344452
[18] Warren Buffett (15 October 1998), lecture at the University of Florida business school (on Long-Term Capital Management).
[19] Robert L. Axtell (2001), “Zipf Distribution of U.S. Firm Sizes,” Science 293(5536): 1818–1820. https://doi.org/10.1126/science.1062081
[20] Pierre Azoulay, Benjamin F. Jones, J. Daniel Kim, Javier Miranda (2020), “Age and High-Growth Entrepreneurship,” American Economic Review: Insights 2(1): 65–82. https://doi.org/10.1257/aeri.20180582
[21] Rajshree Agarwal, Raj Echambadi, April M. Franco, M. B. Sarkar (2004), “Knowledge Transfer through Inheritance: Spin-out Generation, Development, and Survival,” Academy of Management Journal 47(4): 501–522. https://doi.org/10.2307/20159599
[22] Michael Polanyi (1966), The Tacit Dimension, Doubleday.
[23] Colin Camerer, Dan Lovallo (1999), “Overconfidence and Excess Entry: An Experimental Approach,” American Economic Review 89(1): 306–318. https://doi.org/10.1257/aer.89.1.306

Note

This is an opinion article. All figures are drawn from the public statistics listed, with their data grade stated; the gross-margin ladder is based on the Ministry of Finance’s industry profit standards and involves no financial data of any specific company. This article does not constitute investment, entrepreneurship, tax or legal advice.

Corrections

Revised 2 October 2026: thirteen external references on survivorship bias, ergodicity and international evidence were added along with the FAQ, and the following figures were corrected (the conclusions are unchanged):
① In Table 3, the expected value of entrepreneurship was originally NT$26.02–26.46 million with a net difference of −NT$15.54 to −NT$15.98 million, which did not match the sum of the rows; corrected to NT$25.98–26.35 million and −NT$15.65 to −NT$16.02 million, shown as two columns for 19% and 57%;
② the failure breakdown in Section 6 originally said 3.5 years and 16.5 years, which did not multiply back to the amounts; now shown as 3.54 years and 16.46 years;
③ the share of expected value from the NT$100-million branch in Section 8 was originally 2.8%; corrected to 0.9%–3.4%;
④ “53.4% go to zero” in Section 9 was on a tax-registration basis; corrected to the operating basis used in this article, 56.0%.
Further corrections the same day: ⑤ the bracket in the Section 6 failure breakdown originally read “NT$10 million invested × 20% recovery rate”; changed to “NT$10 million invested, a 20% recovery rate, only NT$2 million back”; ⑥ the Section 6 baseline originally read “salary NT$2 million, NT$42 million over twenty years,” which do not multiply back; now noted as a model setting; ⑦ the opportunity cost in Section 8 originally read “median employee NT$720,000”; changed to “ordinary employee NT$720,000,” noted as the opportunity cost set in the model and on a different basis from the DGBAS median of NT$546,000 in Section 5, with the graduate-degree NT$1.13 million noted as an average; ⑧ the count of data items at the start of the appendix was recounted against the full table, and the statement about “the probability side” was corrected (the 60% take-home rate is also an industry estimate); ⑨ the timing of Buffett’s remarks on Long-Term Capital Management was corrected from “when it collapsed” to “when it came close to collapse”; ⑩ FAQ question 1 now mentions the census calibration; ⑪ the Section 8 study by Agarwal and colleagues originally said “the stronger the parent’s technical and market capabilities, the longer the new firm lived”; corrected to say that technical capability raised survival, while market-pioneering capability had no significant effect; ⑫ the term “tacit knowledge” comes from Polanyi in 1958; The Tacit Dimension (1966) is where the line “we know more than we can tell” appears.

License

This work is published by JIAHON Group. All rights reserved. Quotation and redistribution are permitted with proper attribution and a link to jiahon.com.tw. Commercial use—including but not limited to paid courses, reports, or consulting services incorporating this content or its frameworks (including the “lottery ruler,” “gross-margin ladder” and “ticket” models)—requires prior written authorization from JIAHON. For commercial licensing enquiries, please contact us through the website.

Back to top